Use reasonable initial values (still pretty fast, and now good results)

aflaxman · May 12, 2011 · e8c376b · e8c376b
1 parent 49dea95
commit e8c376b
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 3 deletions.
diff --git a/README b/README
@@ -150,5  150,10 @@ I'm not keeping careful track of if the results change as I modify the
 step methods, which is probably not a recommended practice.  But it is
 fun to see the methods get faster and faster as I group more and more
 in the AM step method.  (43s wall time for M.B and M.U in one AM,
-commit here)
 commit here; 32s wall time for all stochs together, but the results
 are not pretty.)
 
 Sometimes I've had good luck with choosing initial values for the MCMC
 by maximizing the posterior or something related to it.  Here is code
 to do that, which doesn't take very long (36s wall time, and
 respectible looking results are back).
diff --git a/analysis1.py b/analysis1.py
@@ -18,14  18,14 @@
 Z = Z[0:240:1,0:10:1]
 
 ## Priors
-var_u = pymc.Gamma('var_u', alpha=1, beta=1)
 var_u = pymc.Gamma('var_u', alpha=1, beta=1, value=1.)
 tau_u = pymc.Lambda('tau_u', lambda v=var_u: v**-1)
 
 B = pymc.Normal('B', mu=[0, 0], tau=10000**-1)
 
 U = pymc.Normal('u', mu=0, tau=tau_u, value=np.zeros(10))
 
-var_e1 = pymc.Uniform('var_e1', lower=0, upper=100)
 var_e1 = pymc.Uniform('var_e1', lower=0, upper=100, value=1.)
 tau_e1 = pymc.Lambda('tau_e1', lambda v=var_e1: v**-1)
 
 @pymc.deterministic

diff --git a/run_analysis1.py b/run_analysis1.py
@@ -4,6  4,12 @@
 from pylab import hist, show
 from pymc import Matplot
 
 # Generate reasonable initial values by maximizing posterior with
 # variances held fixed
 M = pymc.MAP([analysis1.B, analysis1.U, analysis1.y_i])
 M.fit(method='fmin_powell', verbose=1)
 
 # Sample from full posterior distribution
 M = pymc.MCMC(analysis1)
 M.use_step_method(pymc.AdaptiveMetropolis, [M.B, M.U, M.var_e1, M.var_u])
 M.sample(iter=100000, burn=50000, thin=10, verbose=2)